future of Complex Systems Science. Even if the scientific method requires a reductionist approach, that it is not always justified in the case of complex systems, nevertheless it could point out the ‘true’ control parameters and the relevant observables to predict criticalities. Furthermore, it could suggest strategies to perform a control theory approach or to optimize in a robust and generic way the system performance in the framework of ABMs. In the specific problem of ATM, the predictability and the efficiency of the proposed ABM can be significantly improved by the results of a complex physics approach. ABMs have taken advantage from the continuously increasing of the computer performances to cope with large complex systems as the transportation systems.
Nowadays they are one of the most promising approach to simulate complex socio-technological systems to understand the emergent properties of the considered systems related to the cognitive behavior of its components. In the definition and application of an ABM one has to face the following problems:
1. identify the all different agents that have to be modeled to simulate the considered system;
2. define the algorithmic procedures at the base of the agent behavioral strategies;
3. define the interaction between the cognitive behavior of the agents and the physical dynamics of the system;
4. perform the validation process using suitable empirical observations;
5. analyze the predictability of the ABM when new scenarios are presented.
Only when the previous items are
exhaustively studied, an ABM is a powerful tool ready to assist stakeholders to manage complex systems.”
Agent Based Models (ABMs) are therefore research tools used in several fields to observe, model and foresee scenario simulations of highly interconnected complex systems. Within the field of Air Traffic Management (ATM) research they are used to model present state of very complex socio-technical systems or sub-systems and to provide computer simulated scenarios of future settings.
The final aim is to reach maturity in the ability of describing emergent phenomena in complex systems and use them as a decision-support tools for policy choices concerning the evolution and innovations to be introduced in the air transportation system.
i) To what extent agent based models and multi-agents systems have been implemented in the study of the air traffic management complex system? Similarly, to what extent has existing research been able to demonstrate the value of these tools and what actions are required to move further towards industry adoption of such tools?
One of the two experts has pointed out that the ELSA project has developed agent-based model of air traffic management system operations on a strategic and tactical level.
Agents introduced at the strategic level are Airline Operators (AO) and Network Manager (NM) while at the tactical level agents are Aircraft/Pilots and Air Traffic Controllers (ATCo’s). In the report it is stressed that a two-level model (strategic vs tactical) represents a reasonable
approach because it describes a current way of working and because the outputs from strategic model (agreed flight plans between airlines and Network Manager) should be used as inputs into the tactical layer (actual realisation of flight plans).
The ELSA model is of modular structure (contains few modules: flight list, conflict detection, conflict resolution, directs,
shocks, multi-sector) enabling decision makers (of various kinds) to perform experiments in different areas of interest.
According to the external experts the ELSA project has dealt with the first three points mentioned above, whereas further research work is necessary to accomplish the last two tasks. Moreover, it seems that from both strategic and tactical point of view no more agents are necessary, i.e. adequate agents are chosen to represent operations.
Although experiments at the strategic level provide some interesting results, one of the experts highlights some issues related to the interactions amongst agents.
In the view of the external experts, the CASSIOPEIA project is a MAS, whose goal is to reproduce in silico the cognitive behavior of agents that rule the ATM providing both an interface with dynamical databases and the possibility to visualize and to interpret the simulation results. In the present state, the CASSIOPEIA MAS has mainly considered the aircraft agent, the airline agent and the airport agent to perform virtual reality experiments, which simulate different scenarios in the airline flight planning that could be relevant in the future when the air traffic demand is expected to increase. These scenarios are not directly connected to criticalities or congested states of transportation systems, but aim to find an optimized management of air traffic that minimizes the environmental impact, the economic cost of the delays. In order to enable wider usability of simulation platform, agents are defined on three levels:
general (highest level), domain-specific and case-specific (lowest level). This approach is very beneficial because it allows a broad range of applications. Usability of software platform has been demonstrated through three different case studies, each of which has a different scope and agents involved.
The experts suggest that the model shows the possibility of simulating the very complex
interaction processes, which presides over the air traffic at departures and arrival at the airports. The cores of the model are aircraft agents and the airline agents that exchange continuously information with airports and the agents, which manage the arrivals and departures. One of experts suggests the introduction of new types of agents: the Network Manager (NM) and Air Navigation Service Providers (ANSP).
In the case of the work done by B. Monechi, the external experts noticed that ABM has been implemented at the level of Air Traffic Controllers (ATCo’s) and conflict resolution (CR) behavior only. Others agents are not taken into account explicitly,
e.g. airlines or pilots behavior during CR or their acceptance of trajectory
changes. Related to ATCo’s behavior during CR only horizontal resolution strategies are analyzed and modeled although vertical CR are more efficient and usually applied by ATCo’s in current day operations.
As a result, one of the experts highlights that other agents should also be be included in the model as well as traffic flow management strategies applied by ATCo’s and coordination procedure between ATCo’s of adjacent sectors (which is simple modeled).
The main results of the Monechi’s project consists in showing the existence of a generic scaling law, which relates the average flow on the air traffic network and the number of unsolved conflicts
when the flow overcomes a certain threshold (the congestion threshold). The key point is that the exponent which characterizes the scaling law seems to be universal
(i.e. independent from the details of the air traffic network considered), but both the congestion threshold and the exponent depend on the local strategies adopted by air controllers to solve the conflicts. These laws could not only constitute a useful tool to forecast and to control the congestion transitions in an air traffic network, but
also define utility functions in optimizing procedures of the network structure.
In conclusion, the experts’ evaluation suggests that the considered projects have provided valuable contribution to the study of the ATM, and have successfully provided a sound proof of concept that ABMs are a fruitful tool to study this relevant socio-technical complex system.
Each expert has focused his reports on specific aspects, so that further work can be pointed out to generalize these models in order to have a full picture of the system and in order to move the models from the proof of concept state to a state of an industrial product.
ii) What are the different actors of Air Traffic Management that can be most fruitfully modeled by ABMs and MASs?
According to the external experts, the ELSA project might benefit from considering Airline Alliances at the strategic layer as well as Airports needs at the tactical level. Moreover, it might be beneficial to also consider the economic impact of the Network Manager choices on the competing companies.
Also for the CASSIOPEIA project, the experts suggest that the introduction of other
agents should be beneficial. In this case one of them suggests the introduction of the Network Manager and Air Navigation Service Providers/ Air Traffic Controllers into the model.
In the work of B. Monechi the experts deem necessary to model airline/pilot behavior as well as airline trajectory negotiation with Network Manager and local ATC, as well as collecting data and modeling particular situations where the air traffic system has been close to a critical state due to failures in the air traffic network.
In conclusion, it seems that in general the choice (i) of the agents used to model the
ATM and (ii) of the mechanisms by which they interact has been done in a reasonable way, although more agents and interactions should be included in order to enhance the capability of the models to explain the ATM system.
iii) To what extent agent based models and multi-agents systems have shown a maturity level such to be used as part of decision-support tools MASs?
According to the external experts, the ELSA project is of higher maturity and very close to industrial application as a decision-support tool. This model could be certainly a valid support to manage the air traffic in normal condition, reducing the workload of air-traffic controllers and the possible human error.
As for the CASSIOPEIA project, the experts suggest that the presented model provides good starting point for end users to build the models suitable for their problems. The platform is of middle to higher maturity and easily could be used as a decision-support tool, even if a validation process in specific cases is still necessary to quantify the predictability of the model.
The external experts instead think that the work of B. Monechi is not mature enough to be accepted by the industry, However one should consider that the developed model is part of a PhD thesis.
In conclusion, the two projects presented for the evaluation can be considered close to industrial applications.
iv) What are the aspects of the air traffic management system that are worthy of further investigation by making use of agent based Models and multi-agents systems?
According to the external experts, the ELSA project might benefit from considering
• dynamic sectorisation of the airspace,
• the exploitation of different strategies for air traffic flow management,
• a better comparison with standard KPIs,
• the implementation of different economic strategies of the airlines at the strategic layer,
• the implementation of learning strategies especially at the tactical layer.
As for the CASSIOPEIA project, the experts suggest that the project might benefit from
• modeling the Network Manager behavior,
• modeling the ANSP/ATCO activities related to airlines and airports operations,
• a better comparison with standard KPIs,
• modeling of the passengers needs.
In the case of the B. Monechi work, the external experts suggest that it is necessary
• to model airline/pilot behavior during trajectory negotiation process at strategic level and trajectory change process at tactical level,
• modeling of coordination between ATCO’s from adjacent sectors,
• modeling the Network Manager as additional agent interacting with local ATCOs.
Therefore, as mentioned above, there is room for introducing in the models more agents and interactions in order to enhance the capability of the models to explain the ATM system.
According to the external experts, the ELSA project is showing huge potential to be used as a decision-support tool. Its maturity is high which is bringing this model closer to industrial applications. Some further enhancements could be made by adding
some new agents in order to model a wider range of empirical facts. The algorithmic strategies at the roots of the ATM have been exhaustively analyzed in the ELSA project by considering many different real and artificial scenarios, even if some issues need further investigations.
As for the CASSIOPEIA project, the
external experts think that developed ABM platform has huge potential to be used as a decision-support tool provided that further improvements are implemented, especially with the inclusion of new agents and the exploitation of solutions which better take into account the heterogeneity of the different actors involved.
According to the external experts, the work of B. Monechi is showing potential to be used as a decision-support tool provided that further improvements are implemented Those improvements should see the inclusion of new agents as well as some operational procedures currently used or foreseen in future SESAR operational scenario. Moreover, the improvements should also regard the understanding of the physical aspects of congestion transition in transportation systems.
In conclusion, the judgment of the external experts on the way the application of Agent Based Models to the understanding of the ATM socio-technical complex system has been pursued within the Complex World project seems to be positive. Further work is still needed for a better understanding of many stylized facts in ATM, however, and more importantly, ABM seem to have the capability of addressing the open questions.
In particular, it would be beneficial a validation phase in which the developed ABMs/MASs support the decisions of the controllers to see to which extent they prove their capability to manage efficiently a large ensemble of realistic scenarios.
There are many scenarios in ATM where uncertainty plays an important role.
Examples of these include scheduling of arrivals/departures, routing around adverse weather, trajectory prediction, conflict resolution, and flow management. In the past, most integrated decision-support tools (DST) that have been developed to help manage these scenarios commonly neglect uncertainty. However, including the effect of uncertainty in DSTs might help to improve their efficiency, thus benefiting the ATM system.
There is not an unique way to include uncertainty in a DST. Possibilities range from considering the worst case scenario, buffers (such as intervals or confidence ellipsoids), Monte Carlo simulations, or more detailed statistical models.
However, there are many challenges in including uncertainty in a DST. For instance, it is not clear what type of statistical models should be used to realistically capture uncertainty. There is also a trade-off between robustness and performance:
if one tries to accommodate too high levels of uncertainty, it might lead to excessive conservativeness in DST solutions. In addition, while in a deterministic setting an optimal solution is easy to define, this notion is not totally clear in an uncertain environment.
i) Are the different sources of uncertainty in ATM properly identified and
characterized? Is there any important uncertainty source that has been left out?
It is clear that different scenarios would require different modelling of uncertainty sources. The research under review
investigates widely different examples with significantly varying uncertainty modelling.
For instance, the ATFM is formulated including uncertainty in the future capacity of route and airport, which are assumed to be given in terms of static probabilities associated with capacity. However it is unclear how realistic this representation of uncertainty is or how the prescribed probabilities in capacity are determined in practice. The arrival scheduling problem is formulated as an optimisation problem on a graph, where the key parameters are the arrival and take off times; the authors consider uncertainty in these variables. The trajectory predictor computes uncertainty in estimated arrival time (ETA) due to uncertainties in the wind forecast, initial altitude errors, and errors due to the flight navigation system.
All these examples leave some important sources out of the picture, without proper justification in some cases. In
route-planning problems, weather is an important factor which is oftentimes left out. Other relevant sources of uncertainty which are difficult to model and frequently neglected include aircraft malfunction/servicing, pilot/
crew availability, departure/arrival delays, airborne holding, etc. TPs frequently do not consider uncertainty due to passenger boarding, gate delays, airport congestions which may significantly affect predicted ETA. On the other hand results from the literature show that in the context of runway optimisation, factors such weather, en-route flight delays, propagated delays from other airports, maintenance delays, etc., all manifest as uncertain delays in take off and arrival and do not need to be included explicitly.